1 Heisenberg Representation

What we have been dealing with so far is called the Schr odinger representation. In
this representation, operators are constants and all the time dependence is carried by the
states. We have
|(t) >= U (t)|(0) >
where
i
U (t) = exp( Ht)
h

A matrix element of an operator is then

< (t)|O|(t) >

where O is an operator constructed out of position and momentum operators. To contrast

the Schrodinger representation with the Heisenberg representation (to be introduced
shortly) we will put a subscript on operators in the Schrodinger representation, so we
have XS , PS , and OS . We may then write our matrix element as

< (t)|OS |(t) >=< (0)|U (t)OS U (t)|(0) >

The Heisenberg representation uses time dependent operators and constant in time states.
We dene the Heisenberg operator by

OH (t) = U (t)OS U (t)

The two representations are clearly completely equivalent, and it is a matter of conve-
nience which one is used in a given problem.
Once we have dened Heisenberg operators, we may study their equations of motion
and compare to the corresponding classical equations of motion. We will carry through
the discussion for a d = 1 system where the x coordinate ranges to +. We start
with the classical discussion. There is a Lagrangian
m dx 2
L= ( ) V (x),
2 dt
with generalized momentum
L
p=
x
where x = dx/dt. From the Lagrangian we construct a classical Hamiltonian,

p2
Hc = px L = + V (x)
2m
From the classical Hamiltonian, we get classical equations of motion,
Hc p Hc V
x = = p = =
p m x x
Quantizing the system involves introducing operators XS , and PS , which satisfy

[XS , PS ] = i
h.

The quantum Hamiltonian is

PS PS
H= + V (XS ),
2m
and the Heisenberg operators
i i
XH (t) = exp( Ht)XS exp( Ht),
h
h

and
i i
PH (t) = exp( Ht)PS exp( Ht).
h
h

The quantum or Heisenberg equations of motion are
d i
XH (t) = [H, XH (t)]
dt h

and
d i
PH (t) = [H, PH (t)].
dt h

Note that H being constant is the same in Heisenberg and Schrodinger representations,
PS PS PH PH
H= + V (XS ) = + V (XH )
2m 2m
Now consider the Heisenberg equation for XH . We have
d i i
XH (t) = [H, XH (t)] = U (t)[H, XS ]U (t)
dt h
h

Now
PS PS PS PS
[H, XS ] = [ + V (XS ), XS ] = [ , XS ] + [V (XS ), XS ]
2m 2m
Since every operator commutes with itself,

[V (XS ), XS ] = 0,

and we are left with

PS PS 1 h
PS
[ , XS ] = (PS PS XS XS PS PS ) = ,
2m 2m i m
where we used [XS , PS ] = i
h. Putting the pieces together, we have
d i h
PS PH (t)
XH (t) = U (t) U (t) = (1)
dt h
i m m
which is the same as the classical result for dx/dt.
Now let us get the Heisenberg equation of motion for PH . We have
d i i
PH (t) = [H, PH (t)] = U [V (XS ), PS ]U,
dt h
h

where we used the fact that PS commutes with itself. To get [V (XS ), PS ], we rst take
a matrix element and write

< x|[V (XS ), PS ]| >=< x|(V (XS )PS PS V (XS ))| >
h
h

= V (x) x < x| > (x V (x) < x| >)
i i
h
h
V (XS )
= (x V (x)) < x| >= < x| | >
i i XS
So we have
h
V (XS )
< x|[H, PS ]| >= < x| | >
i XS
Since < x| and | > were arbitrary, we can write the operator equation
V (XS )
[H, PS ] =
XS
Returning to the equation for dPH /dt, we now have
d V (XS ) V (XH )
PH (t) = U (t) U (t) = , (2)
dt XS XH
which again is the quantum version of the classical equation for dp/dt. It is generally true
in a quantum system that the Heisenberg equations of motion for operators agree with the
corresponding classical equations. An important example is Maxwells equations. These
remain true quantum mechanically, with the elds and vector potential now quantum
(eld) operators.

Application to Harmonic Oscillator In this section, we will look at the Heisenberg

equations for a harmonic oscillator. The notation in this section will be O(t) for a
Heisenberg operator, and just O for a Schrodinger operator. In terms of the notation of
the previous section we have OS = O, and OH (t) = O(t). Of course we have O(0) = O.
The Hamiltonian for the oscillator is
PP m02 X 2
H= + , (3)
2m 2
where 0 is the natural frequency of the oscillator. The equations of motions for the
Heisenberg operators are as follows,
dX(t) P (t) dP (t)
= , = m02 X(t)
dt m dt
As always, the Heisenberg equations for operators are the same as the classical equations
of motion. Taking a second time derivative, we have
d2 X(t) 1 dP (t)
2
= = 02 X(t)
dt m dt
This is a dierential equation for X(t). We can certainly solve it as a linear combination
of sin 0 t and cos 0 t. We can write
X(t) = A cos 0 t + B sin 0 t,
where A, B are operators which are independent of t. At t = 0, we must have X(0) = X,
so we get that
A = X.
Also from Eq.(3), we have
dX(t) P
|t=0 = = 0 B,
dt m
so
P
B= .
m0
We nally have
P
X(t) = X cos 0 t + sin 0 t. (4)
m0
Similar steps lead to
P (t) = P cos 0 t = m0 X sin 0 t. (5)

Matrix Elements and Energy Levels The expressions for X(t), P (t) look so
much like their classical counterparts, it might seem unlikely that they contain informa-
tion about energy levels and matrix elements. As will be seen, they in fact contain a
large amount of such information. Suppose the Hamiltonian has an eigenstate |n > with
energy En and another eigenstate |n > with energy En . Let us write out the matrix
element of X(t). We have
iHt iHt i(En En )t)
< n |X(t)|n >=< n | exp( )X exp( )|n >= exp( < n |X|n > .
h
h
h

(6)
This is the matrix element of the left hand side of Eq.(4). Using exponential forms for
sin and cos we have for the matrix element of the right hand side,
( )
1 i0 t P P
e (< n |X|n > i < n | |n >) + ei0 t (< n |X|n > +i < n | |n >) . (7)
2 m0 m0
Eqs.(6) and (7) must match in detail. Let us start with the case En > En . Then com-
paring time dependent factors on both sides, the coecient of exp(i0 t) must vanish,
since there is no term of that form in Eq.(6). This gives
P
< n |X|n > +i < n | |n >= 0. (8)
m0
Now matching the coecient of exp(i0 t, we also must have
i(En En )t)
exp( ) = exp(i0 t)
h

This result implies that if < n |X|n >= 0, then

En = En + h
0 .

Let us assume such a state exists. Now apply the same argument to the matrix element

< n |X(t)|n > .

We nd again, if En > En that

En = En + h
0 ,

provided only that < n |X|n >= 0. Using this argument repeatedly, we nd a sequence
of levels En , En + h
0 , En + 2
h0 , . . . , En + l
h0 , with no upper limit on l, provided only
that the matrix element of X between a state and the next one higher up is non-vanishing.
We will return to the question of which matrix elements of X are non-vanishing later
in this section. Before doing that, we explore the case where En < En . Returning to
Eqs.(6) and (7), this time the coecient of exp(i0 )t must vanish, giving
P
< n |X|n > i < n | |n >= 0. (9)
m0
The coecient of exp(i0 t) must match on both sides, so we get

En = En h
0 ,

0 lower, provided that < n |X|n >= 0.

so we have found a state with energy one unit of h
At rst sight it would seem that this process could be repeated indenitely. However, this
would eventually result in the energy eigenvalue going negative. But this is impossible.
Consider the expected value of H in a state | > . Writing this out in Hilbert space
notation instead of Dirac notation, we have
1 1 1 1
(, H) = (, P P ) + m02 (, XX) = (P , P ) + m02 (X, X). (10)
2m 2 2m 2
Each term on the right side of Eq.(10) is certainly positive, regardless of the particular
state. The consequence is that H can have no negative eigenvalues. This means that
the argument we have been using cannot generate an innite sequence of levels below
a given one, but it is allowed to generated an innite sequence above a given one. The
consequence is that there must be a lowest state, which we denote as |0 > . It is then
natural to designate the sequence of levels upward from |0 > as |n >, where

En = E0 + n
h0 ,
where n is any positive integer.
Finally, let us explore the question of which matrix elements of X and P are non-
vanishing. We start with the lowest state, |0 > . By the discussion just given, both X
and P, acting on |0 >, must lead to a state of higher energy. Let us take matrix elements
between |0 > and |1 > . This puts us in the case covered by Eq.(8) so we have
P
< 1(|X + i )|0 >= 0. (11)
m0
This relates matrix elements of X and P. To get a more powerful result, we consider
matrix elements between |0 > and |n > for n > 1. It is easy to see that all such ma-
trix elements of X and P must vanish, since the matrix elements < n|X(t)|0 > and
< n|P (t)|0 > have time dependence exp(in0 )t). But from Eqs.(4) and (5), the only
allowed frequency for the matrix element of a state higher in energy is 0 . ( This result
holds more generally, namely the matrix elements of X and P are only non-vanishing
between a state and the next one up or down in energy. ) Now consider the expression
P
(X + i )|0 >
m0
From Eq.(11) it has no matrix element to |1 >, and since the matrix elements of both X
and P vanish for any higher state, we have that
P
(X + i )|0 >= 0 (12)
m0
We can apply any operator to Eq.(12) and still get a vanishing result, so we also have
P P
(X i )(X + i )|0 >= 0 (13)
m0 m0
Multiplying this out we get
{ }
PP i
XX + 2
+ [X, P ] |0 >= 0 (14)
(m0 ) m0
Supplying a factor m02 /2 and using [X, P ] = i
h, we nally have
h
0
(H )|0 >= 0 (15)
2
or
h
0
E0 =
2
To summarize, we have found the entire sequence of energy levels of the harmonic oscil-
lator. It is also easy to show that X and P only have matrix elements between |n > and
|n 1 >, except for n = 0, where the only non vanishing matrix elements are to |1 > .
A little more work of the same sort done above will deliver the actual values of these
matrix elements. This exercise shows that Heisenberg methods can be quite powerful in
certain cases.